The Arch-Global E-Commerce team develops large-scale recommendation, matching, and ranking algorithms for creator-commerce platforms, supports multiple teams such as Creator Marketplace, Creator Shop, and TikTok consumer experiences. We leverage machine learning and data-driven optimization to connect sellers, creators, and products more effectively, increase content-commerce supply, and accelerate e-commerce GMV growth.
You will own the overall algorithm infrastructure roadmap for Creator Agent. Work in close partnership with the algorithm team to ensure cutting-edge agent capabilities can be delivered to the creator business.
Responsibilities:
- Lead the architectural development of core Agent algorithm capabilities, including but not limited to:
- Agent orchestration framework: Build agent orchestration capabilities supporting complex business logic, based on LangGraph or in-house frameworks.
- Agentic Search: Build intelligent retrieval architecture tailored to creator scenarios, enabling the Agent to proactively and iteratively gather information from product, creator, and content corpora.
- Hierarchical memory systems: Design short-term, long-term, and episodic memory mechanisms, providing the algorithm team with foundational capabilities for personalized creator understanding.
- Algorithm tuning infrastructure: Provide efficient training, evaluation, and iteration infrastructure for Agent RL, Memory RL, SFT, and additional frontier optimization directions (see below).
- Continuously track and bring frontier Agent optimization directions into the team, including but not limited to:
- Test-time / inference-time optimization (self-refine, reflection, tree search, process reward model–guided reasoning, etc.);
- Tool use optimization (tool-use SFT, tool-use trajectory RL, tool selection optimization);
- Multi-agent collaboration and deliberation;
- Automated prompt / workflow optimization (e.g., DSPy, TextGrad — "gradient-style" optimization of prompts and workflows);
- Agent distillation into smaller, more efficient models;
- Agent evaluation and reward modeling (LLM-as-Judge, PRMs, Agent benchmark design, etc.);
- And, based on team and business realities, judge which directions are worth investing in and translate them into team capabilities.
- Track the latest Agent architectures from OpenAI, Anthropic, and others, and adapt them deeply to our creator business.
- Partner deeply with the Algorithm team — understands the real needs of algorithm iteration, and ensures that algorithm infrastructure accelerates rather than bottlenecks algorithmic innovation.
- Develop deep understanding of creators as a B2B user group; translate business insights into algorithm infrastructure decisions.
Requirements
Minimum Qualifications:
- Deep understanding of the Agent technical stack — familiarity with the architectural approaches of frontier Agent practices such as OpenAI SDK and Claude Code, and a clear point of view on the capability boundaries and algorithmic challenges of Agents.
- Hands-on experience with LangGraph (or equivalent frameworks) for building production-grade, domain-customized agents.
- Systematic understanding of Agent optimization — familiarity with the foundational directions (Agent RL, hierarchical memory and Memory RL, SFT), plus hands-on practice or deep familiarity with at least 2-3 frontier directions (e.g., test-time optimization, tool-use optimization, multi-agent collaboration, automated prompt/workflow optimization, Agent distillation, Agent evaluation).
- Familiarity with Agentic Search design and implementation, with a clear understanding of the paradigm shift from traditional retrieval to agent-driven retrieval.
- Deep understanding of B2B / ToB businesses — able to reason about algorithm infrastructure from B2B-specific angles (user workflows, ROI, explainability, controllability) rather than directly applying consumer-product intuitions.
- Technical judgment and forward-looking perspective — able to identify directions with real business value in a rapidly evolving Agent landscape, and bring fresh thinking to the team.
- Outstanding cross-team collaboration skills — able to build a deep partnership with the product, algorithm and other cooperation engineer teams and drive complex algorithm projects to high-quality delivery.
Preferred Qualifications:
- End-to-end experience building LLM-powered (especially Agent) algorithm infrastructure from 0 to 1.
- Deep hands-on experience with RL / RLHF / DPO / GRPO and other LLM alignment techniques.
- Hands-on exploration or published work in test-time scaling, process reward models, or Agent self-improvement.
- Practical experience with automated prompt / workflow optimization (DSPy, TextGrad, etc.).
- Experience with algorithm infrastructure for content generation (copywriting, scripts, etc.).
- Experience with B2B / SaaS / CRM businesses, either on the algorithm or infrastructure side.
- Experience with international / multilingual / multi-region products.